If you’ve been hurt by online fraud, you know firsthand how frustrating and damaging it can be to a business. And, unfortunately, you’re not alone. Merchants lose an estimated $3.5B in online revenue to fraud annually. Fraud can take the form of chargebacks, fake account sign ups, stolen credit cards, identity theft, and more.
Thanks to big data and Sift Science’s machine learning technology, we’ve found numerous red flags that point us straight to potential online criminals. While determining a good order from a bad one is rarely simple, data — especially when used to train machine learning systems — can help streamline the process and make your decisions more accurate. To kickstart your fraud detecting, Visa’s “Merchant Guide to Greater Fraud Control” reveals 10 indicators to consider:
- First-time shoppers: Criminals constantly search for new merchants to bamboozle, creating new shopper accounts on merchant sites.
- Bulk orders: Criminals try to maximize their order sizes because stolen credit cards have unpredictable life spans.
- Orders that include variations of the same item: Buying multiple high-value goods — such as luxury watches, fancy handbags, or expensive tennis shoes — is suspicious.
- “Rush” or “overnight” shipping: Criminals want to profit off their stolen goods ASAP.
- International shipping: Fraudulent transactions are sometimes shipped to criminals outside of the merchant’s country.
- Inconsistencies: Differences in billing and shipping addresses, telephone area codes and zip codes, non-legitimate email addresses, and orders at odd times of day are to be considered potentially fraudulent.
- Multiple credit cards shipping to the same address: This could indicate a single user with a bundle of stolen cards.
- Many transactions with the same billing address, but multiple shipping addresses: Organized criminal activity, originating from a single hub.
- More than two cards used from the same IP address: A greater-than-normal number of cards could indicate fraud.
- Orders from emails that are not linked to any billing relationships: Bad users may create email addresses purely to purchase.
Unfortunately, this list is not exhaustive. Criminals have become increasingly creative and unpredictable, making fraud difficult to detect if you don’t know where to look. Some indicators (“signals”) are more suspicious than others. Did you know that a buyer with multiple billing zip codes within a week is 30 times more likely to be fraudulent than the normal user?
That’s where machine learning and intelligent, adaptable fraud detection comes in.
Dealing with multiple flags and alerts can be annoying and a huge headache. Traditional fraud detection systems produce a 50-80 percent false positive rate — meaning that they incorrectly identify a good user or order as fraud up to 80% of the time!
At Sift Science, we combat fraud using real time machine learning technology, custom-tailored to your business that adapts to fraudulent behavior in milliseconds. We sift through millions of fraud patterns like the ones above and understand how to weigh the value of subtle cues that a traditional rule-based system would miss.
Curious about machine learning? Check out our article on How Our Machine Learning Works.
Want to learn more about online fraud? Check out our free fraud education portal.
Stay tuned for our next post in this series on “10 Warning Signs of Fraud”. We’ll dive deeper into each of the 10 signals and give you tips and tricks on avoiding the pitfalls of fraud. You can always follow us on Twitter at @siftscience.